Bug#37635: Test read_many_rows_innodb is failing w/ wrong error message (lock_wait_timeout)
The test is supposed to provoke a deadlock, to test a crash of the server in some deadlock scenarios. The problem is that in recent version of MySQL, no deadlock occurs. It is not clear why a deadlock should be expected. One transaction does an insert+delete on primary key value 1 in table t2. The other transaction does an insert on primary key value 123. There seems no reason that these should conflict, so there is no deadlock. (Presumably an earlier version did get a deadlock due to maybe taking excessive locks on the insert+delete of primary key value 1. The table has only one row, so maybe a different locking was used ("Impossible WHERE noticed after reading const tables")) Fix is to have transaction 1 lock PK 1 and transaction 2 lock PK 123. Then have transaction 2 access PK 1 (wait) and transaction 1 access PK 123 (deadlock). It has been checked that this modified test case still crashes the server in the same way prior to fix of the original bug (Bug#24989). mysql-test/include/read_many_rows.inc: Provoke deadlock on same PK value in the two transactions to ensure that the deadlock does occur independently on whatever execution plan is chosen. mysql-test/r/read_many_rows_innodb.result: Result file update.
Showing
Please register or sign in to comment